Status of PAE support?

Status of PAE support?

[Christopher S. Aker]

Hello,

Enabling PAE in xen/include/asm-x86/config.h results in the following boot output:

http://www.theshore.net/~caker/xen/PAE-testing/test1/boot.log

It boots OK if I disable PAE.  I can proide binaries if needed.

Thanks,
-Chris

Re: Status of PAE support?

[Keir Fraser]

On 7 Jun 2005, at 03:23, Christopher S. Aker wrote:

> Hello,
>
> Enabling PAE in xen/include/asm-x86/config.h results in the following 
> boot output:
>
> http://www.theshore.net/~caker/xen/PAE-testing/test1/boot.log
>
> It boots OK if I disable PAE.  I can proide binaries if needed.

You need a PAE-enabled domain0 kernel as well, and our tree isn;t yet 
patched to support PAE. So bootstrapping Xen is as far as you'll get 
right now.

  -- Keir

Re: Status of PAE support?

[Gerd Knorr]

Keir Fraser <Keir.Fraser@cl.cam.ac.uk> writes:

> > http://www.theshore.net/~caker/xen/PAE-testing/test1/boot.log
> >
> > It boots OK if I disable PAE.  I can proide binaries if needed.
> 
> You need a PAE-enabled domain0 kernel as well, and our tree isn;t yet
> patched to support PAE. So bootstrapping Xen is as far as you'll get
> right now.

http://dl.bytesex.org/patches/ has xenlinux patches for you, havn't
found the time yet to update them for after-initial-pae-merge source
tree, so it likely is a bit tricky to get it going at the moment.

Seems the pae patch chunk for common/elf.c got lost on the merge (by
mistake or intentionally?).  I had added a sanity-check there to catch
the case of attempting to boot a non-pae kernel on pae xen (or
visa-versa), so you'll get a clear error message instead of a kernel
crashing very early.  Hmm, something wrong with that?

  Gerd

Index: xen/common/elf.c
===================================================================
--- xen.orig/common/elf.c	2005-05-30 14:09:15.000000000 +0200
+++ xen/common/elf.c	2005-05-30 14:14:30.000000000 +0200
@@ -82,7 +82,20 @@ int parseelfimage(struct domain_setup_in
             printk("ERROR: Xen will only load images built for Xen v3.0\n");
             return -EINVAL;
         }
-
+#if defined(__i386__)
+#ifdef CONFIG_X86_PAE
+        int xen_pae = 1;
+#else
+        int xen_pae = 0;
+#endif
+        int guest_pae = strstr(guestinfo, "PAE=yes") ? 1 : 0;
+        if (xen_pae != guest_pae) {
+            printk("ERROR: PAE mode mismatch (xen=%s,guest=%s)\n",
+                   xen_pae   ? "yes" : "no",
+                   guest_pae ? "yes" : "no");
+            return -EINVAL;
+        }
+#endif
         break;
     }
     if ( guestinfo == NULL )

Re: Status of PAE support?

[Christopher S. Aker]

> > You need a PAE-enabled domain0 kernel as well, and our tree isn;t yet
> > patched to support PAE. So bootstrapping Xen is as far as you'll get
> > right now.
>
> http://dl.bytesex.org/patches/ has xenlinux patches for you, havn't
> found the time yet to update them for after-initial-pae-merge source
> tree, so it likely is a bit tricky to get it going at the moment.

The http://dl.bytesex.org/patches/linux-xen-12/pae-support patch applied to xen0
without any problems, compiled and booted much farther along (into userspace!) but
BUGS in hypercall.h:387.

Complete boot log and binaries here:
http://www.theshore.net/~caker/xen/PAE-testing/test2/

After that, I realized the stuff in the hypercalls patch is likely needed.  I was
able to manually merge most of the stuff from it into include/asm-xen/hypervisor.h,
but got in over my head with the changes required to mm/hypervisor.c.

-Chris

Re: Status of PAE support?

[Gerd Knorr]

> The http://dl.bytesex.org/patches/linux-xen-12/pae-support patch applied to xen0
> without any problems, compiled and booted much farther along (into userspace!) but
> BUGS in hypercall.h:387.

http://dl.bytesex.org/patches/linux-xen-12/series is the apply
order of the patches (on top of vanilla 2.6.11).  You should
apply all patches, except the "hypercall" patch.  The hypercall
changes are not merged yet and will be done in a different way
(see recent discussions on the list).

Without hypercall changes you can't use memory above 4 GB, so
there is no point (yet) in using PAE, except for testing and
bughunting ;)

> Complete boot log and binaries here:
> http://www.theshore.net/~caker/xen/PAE-testing/test2/

Could be the missing "shared-pmd" patch.

  Gerd